> Eventually others will reach the level of GPT-3 and they won't lock it away. What happens then?
The same thing that happened for GPT-2. By that time OpenAI will be already months in making GPT-4. And the bigger the model the harder it is for anyone to replicate, especially open source.
Also OpenAI business model is an hosted API. It's something to have access to a GPT-3 clone weights, quite another thing to use it for inference, work is then needed doing compression and distillation to get acceptable cost/latency.
So all in all, OpenAI has at least a year where they can run their API with 0 competition whatsoever, and after that they should be ready to release the next iteration.
3 comments
[ 2.5 ms ] story [ 17.4 ms ] threadEventually others will reach the level of GPT-3 and they won't lock it away. What happens then?
Maybe we should open it and build tools to detect fakes just like we have pretty good tools to detect deep fakes that even humans can't detect.
The same thing that happened for GPT-2. By that time OpenAI will be already months in making GPT-4. And the bigger the model the harder it is for anyone to replicate, especially open source.
Also OpenAI business model is an hosted API. It's something to have access to a GPT-3 clone weights, quite another thing to use it for inference, work is then needed doing compression and distillation to get acceptable cost/latency.
So all in all, OpenAI has at least a year where they can run their API with 0 competition whatsoever, and after that they should be ready to release the next iteration.